Suggestion: a new function ggcoxadjustedcurves to fix #67

[pbiecek]

In #67 @markdanese proposed to add survival curves for cox model.
Not curves for average values of variables but curves averages over the distribution of variables observed in the data (like here: https://cran.r-project.org/web/packages/survival/vignettes/adjcurve.pdf, or in Marks' code here: https://gist.github.com/markdanese/d1ddfdb2f618373c719bdb444ca34be9)

Suggestion: add a ggcoxadjustedcurves() function, that will plot survival curve for the whole dataset (average over all individual curves) or for subsets generated by a given variable.

In addition, al individual curves may be also added to the plot.

Some examples:
For whole dataset

For selected variables

btw: I am not familiar with data.table, so this function uses dplyr and tidyr

[kassambara]

R CMD check:

checking R code for possible problems ... NOTE
ggcoxadjustedcurves: no visible binding for global variable ‘time’
ggcoxadjustedcurves: no visible binding for global variable ‘value’
ggcoxadjustedcurves: no visible binding for global variable ‘.id’
Undefined global functions or variables:
.id time value
Consider adding
importFrom("stats", "time")
to your NAMESPACE file.

[kassambara]

In the survminer functions eval(fit$call$data) is commonly used to get the model data. This lead to repetitive issues, for example #125 & #9.

It might be better, if we add an extra argument, say data, so that user can provide the model data.

If yes, then we should update ggsurvplot(), ggcoxfunctional() & ggcoxadjustedcurves(). The R code to be added in these functions would look like:

if(missing(data)){
    warning ("...warning message...")
   data <- eval(fit$call$data)
}

What do you think @pbiecek , @MarcinKosinski ?

[kassambara]

An error is generated with the R code below:

library(survival)
library(survminer)

fit <- coxph(Surv(time, status) ~ age + ph.ecog+sex, data =  lung)
ggcoxadjustedcurves(fit)

geom_path: Each group consists of only one observation. Do you need to adjust the group aesthetic?
Warning message:
Removed 185 rows containing missing values (geom_path). 

But, this works:

fit <- coxph(Surv(time, status) ~ age + sex, data =  lung)
ggcoxadjustedcurves(fit)

[pbiecek]

Regarding the eval(fit$call$data)
I had an issue when trying to use load() function with Shiny. Then the data is loaded to an other environment and the ggsurv returns and error 'data not found'.
I've checked these alternatives:
all.vars(formula(fit$call) - returns all variables from formula
colnames(broom::augment(fit)) - returns all variables and mode
broom::augment(fit) - returns an extended dataset, for some functions it may be enough.

But I agree, that the best way is to add additional data= argument.
Maybe this is a place for .get_data() function.

[pbiecek]

Regarding NOTEs in CHECK,
they are caused by non standard evaluation used in dplyr/tidyr.
To deal with them I've used a hack as suggested in http://stackoverflow.com/questions/9439256/how-can-i-handle-r-cmd-check-no-visible-binding-for-global-variable-notes-when (solution 2).

[pbiecek]

The problem with
fit <- coxph(Surv(time, status) ~ age + ph.ecog+sex, data = lung)
was caused by missing values. Now I've added na.rm=TRUE to each mean() function and both curves are calculated properly.

[kassambara]

oh, great!!!